for configuration changes
play

for Configuration Changes Herry Paul Anderson Gerhard Wickler - PowerPoint PPT Presentation

Automated Planning for Configuration Changes Herry Paul Anderson Gerhard Wickler LISA Conference, December 2011 Boston, US Outline Declarative Approach Example: Configuration Problem Solution: Declarative Tool Solution: Our


  1. Automated Planning for Configuration Changes Herry Paul Anderson Gerhard Wickler LISA Conference, December 2011 Boston, US

  2. Outline • Declarative Approach • Example: Configuration Problem – Solution: Declarative Tool – Solution: Our Prototype • Experiment: Cloud-Burst Problem – Demo • Conclusions

  3. Declarative Approach • Most commonly used today • Popular tools: Puppet, Cfengine, Chef, LCFG • Critical shortcomings – Indeterminate order executions of actions – Could violates the system’s constraints

  4. Example: Configuration Problem A B A B Actions? C C Administrator Current State Desired State Constraint: C must always refer to a running server! Running Stopping

  5. Solution: Declarative Tools Desired State Puppet • A.running = false Cfengine Submit • B.running = true • LCFG C.service = B Administrator Implement • Possible sequences of states 1) A.running = false C.service = B B.running = true X 2) C.service = B A.running = false B.running = true X 3) B.running = true A.running = false C.service = B X 4) A.running = false B.running = true C.service = B X 5) C.service = B B.running = true A.running = false X 6) B.running = true C.service = B A.running = false √ • Highly likely producing the wrong sequence!

  6. Solution: Our Prototype • All actions must be orchestrated as a workflow to – achieve the desired state – satisfy the constraints • Method – using Automated Planning technique action Declarative approach: Our Prototype: pre action eff pre : preconditions eff : effects

  7. Solution: Our Prototype (2) Desired State • A.running = false • B.running = true Define • C.service = B Global Constraint Administrator • C.service.running = true Current State • Retrieve A.running = true • B.running = false Monitoring Agent • C.service = A Actions pre start ( server ) eff Define pre stop ( server ) eff Experts, Engineers pre change ( s1, s2, c ) eff

  8. Solution: Our Prototype (3) Desired State Current Actions + Constraints State Database Planner Workflow Execution Agents (ControlTier and Puppet) . . . . Managed Servers

  9. Experiment: Cloud-Burst Problem • Cloud-Burst – Migrate application from private to public cloud – Address spikes in demand • Constraints – No down-time – Reconfigure the firewall – Full migration but not duplication

  10. Experiment: Cloud-Burst Problem Company’s LAN Company’s LAN Cloud Provider WS-A WS-B WS-B WS-A VM-X VM-Y VM-Y VM-X Private Cloud Private Public Actions? Firewall Internet C C Administrator Running Stopping

  11. Demo • http://goo.gl/Qph7F • Cloud-Burst problem

  12. Conclusions • Our prototype – Automatically generate the workflow between any two states – Achieve the desired state – Preserving system’s constraints – Enable autonomic reconfiguration

  13. Acknowledgement • This research is fully supported by a grant from 2010 HP Labs Innovation Research Program (IRP) award

  14. Thank you!

Recommend


More recommend